Traffic Simulation in Wireless Cells

1
Traffic Simulation in Wireless Cells
2
Features of Traffic Measurements

• Many research papers have reported self-similar
  features about network traffics
• LeLa94 first reports the self-similar nature of
  Ethernet traffic
• Panx95 reports the self-similar nature of TCP,
  FTP, Telnet and WWW traffics
• Will97 gives a source-level analysis of LAN
  traffic and brief discussion of the WAN traffic
• JNHT01 gives evidences showing that traffic in
  wireless cells also exhibits self-similar features

3
Observation of Traffic Features in the Simulation

• One observation of the traffic features in our
  simulation regardless of the MAC
• the interested traffic is always between the base
  station and the mobile node.
• As long as the base station is speaking or
  listening, all the other traffic in the cell
  interferes.
• All the existing traffic in the cell will compete
  the air channel with the interested flow in the
  simulation.

4
Observation of Traffic Features in the Simulation

• Illustrations of the Observation
• For infrastructure based 802.11
• Every mobile in the cell can listen the signals
  from the base station.
• Result when base station is transmitting, no
  other mobile in the cell can transmit because
  there is no receiver who can avoid the
  interference from the base station.
• The base station is able to pick up signals from
  any mobile in the cell.
• Result when base station is receiving, no other
  mobile in the cell can receive because any other
  transmitter in the cell will cause interference
  at the base station.

5
Observation of Traffic Features in the Simulation

• Illustrations of the Observation (Contd)
• For CDMA
• All the packets should be delivered to the base
  station before they reach their destination.
• No two CDMA mobiles can talk directly to each
  other, the packets need to be relayed by the base
  station(s).
• Result In a single cell, the interference and
  competition all happened at the base station.

6
Simulation Model

• Based on the observation
• The interference traffic is the aggregated
  traffic of all other mobile nodes in the same
  cell.
• According to existing research in self-similar
  traffic modeling, the aggregated traffic can be
  modeled by Fractal Brownian Motion (FBM) at
  source level.
• Media access competition actually can always be
  detected at the base station in the
  infrastructure based wireless network.

7
Simulation Model

• Proposed simulation model
• Using FBM process to generate background traffic
  at the source level.
• Collecting the number of source requests in a
  specified time period by calculating the number
  of packets generated according to FBN model and
  the interested traffic flow.

8
Simulation Model

• Proposed simulation model (Contd)
• Model the MAC competition of packet transmission
  using a MAC competition simulation process
  running in the base station.
• Different MAC protocol can have different MAC
  competition simulation process.
• Different MAC protocol will calculate different
  delivery time and different data rate for the
  delivered packet.
• After the competition, we only send out the
  successful packets belong to our interested flow
  through the air interface, so it will be
  delivered without collisions and errors in MAC
  layer though it is still vulnerable to BER in the
  physical layer.

9
Simulation Model

• Assumptions needed in the model
• Different classes of traffic exist in the
  network.
• In the same cell, mobiles generating the same
  class of traffic have similar behaviors.
• MAC packets in the cell are of the same length.
• Competition model is totally calculated in the
  specific MAC competition process.

10
Simulation Model

• Interference traffic generation
• We use the FBM process described in Will97
  Theorem 1 to generate the interference traffic

11
Simulation Model

• Interference traffic generation (Contd)
• Formula for slim

12
Simulation Model

• Interference traffic generation (Contd)
• Within a cell, different sets of parameters will
  be used to model the typical individual source of
  each traffic class.
• The aggregated traffic of each traffic class will
  be modeled by theorem1 in previous slide.
• The sum of all different traffic classes compose
  the final offered load for the cell.

13
Simulation Model

• Parameters setting
• Offered traffic load
• For WLANKEW00
• Theoretic upper bound around 83 to 86
• Heavy load 75
• Mediate load 30-50
• Low load under 30
• For CDMA
• Throughput is limited by the power and
  interference level only.
• Under that limitation, can be decided randomly.

14
Simulation Model

• Note spatial and temporal correlation of offered
  traffic load
• We decide to model the offered load according to
  the following assumptions
• There isnt tight spatial correlation on offered
  traffic load among different cells. Thus, we can
  randomly decide the offered load in a new cell
  without referring to the offered load in the
  neighboring cells.
• Theres temporal correlation on the offered
  traffic load within a single cell. Thus,
  typically the offered traffic load will not
  change a lot during the time a mobile is in a
  cell.

15
Simulation Model

• Parameters setting (Contd)
• Number of users
• Randomly decided the number of users in the cell.
• Define three classes of traffic
• TCP/FTP data traffic
• Audio/video traffic
• WWW Internet traffic
• For each traffic class, assign a percentage of
  users.
• For WLAN
• 50 data traffic, 30 web/Internet traffic, 20
  audio/video traffic
• For CDMA
• 75 audio traffic, 15 web/Internet traffic, 10
  data traffic

16
Simulation Model

• Parameters setting (Contd)
• ON/OFF Pareto model parameters
• the mean rate can be decided by the number of
  users and the offered load.
• Each different class of traffic will have a
  typical set of Pareto parameters (k, a).
• Based on the formulas described in theorem1,
  generate the simulated traffic for each class and
  sum the results up.

17
Simulation Model

• WLAN MAC algorithm
• Each station will hold the packets until it
  senses the media is free.
• It calculates a random backoff time according to
  the maximum number of contention window slots.
• It waits for a fixed time interval plus the
  random backoff time slots to start its
  transmission of RTS message.
• Every station receiving the RTS message will set
  a time period to be not free according to the
  time flag the RTS message carries.

18
Simulation Model

• WLAN MAC algorithm (Contd)
• The specified receiver in the RTS message will
  send out a CTS message to confirm the reservation
  of the air channel for a certain period.
• Every station receives the CTS message will
  adjust the time period to indicate the busy time
  of the air channel.
• If an RTS or CTS is received before the station
  issued its own RTS, the transmission will be
  postponed.
• Next time when the station senses the air channel
  free after the indicated busy time period, it
  wait for the fixed time interval plus the
  remaining time from the previous random backoff
  time.

19
Simulation Model

• WLAN MAC algorithm (Contd)
• If collision of RTS messages occurs, both
  stations will recalculate a random backoff time
  and be postponed.
• The data packets have a time-out value associated
  with them and will be discarded after the time
  out.

20
Simulation Model

• Illustration of WLAN MAC
• Priorities
• defined through different inter frame spaces
• no guaranteed, hard priorities
• SIFS (Short Inter Frame Spacing)
• highest priority, for ACK, CTS, polling response
• PIFS (PCF IFS)
• medium priority, for time-bounded service using
  PCF
• DIFS (DCF, Distributed Coordination Function IFS)
• lowest priority, for asynchronous data service

DIFS
DIFS
PIFS
SIFS
medium busy
next frame
contention
t
direct access if medium is free ? DIFS
21
Simulation Model

• Illustration of WLAN MAC
• Sending unicast packets
• station can send RTS with reservation parameter
  after waiting for DIFS (reservation determines
  amount of time the data packet needs the medium)
• acknowledgement via CTS after SIFS by receiver
  (if ready to receive)
• sender can now send data at once, acknowledgement
  via ACK
• other stations store medium reservations
  distributed via RTS and CTS

DIFS
data
RTS
sender
SIFS
SIFS
SIFS
ACK
CTS
receiver
DIFS
NAV (RTS)
data
other stations
NAV (CTS)
t
defer access
contention
22
Simulation Model

• MAC competition model for WLAN
• For a fixed length of time period, calculate the
  arriving interference traffic by using FBM model
  in the previous slide.
• Set the length of the time period to be the
  following
• Let S be the length of contention window in
  number of slots, P the time to transmit the fixed
  length packets, W the average backoff slots for
  each packet, then
• Randomly give each packet a backoff time
  according to the length of contention window and
  schedule them into different time slots in the
  time period.
• Mark the virtually generated packet as sent
  when the scheduled time arrives and theres no
  collision.

23
Simulation Model

• MAC competition model for WLAN (Contd)
• Two packets scheduled with the same backoff time
  is considered to collide with each other and both
  are postponed with a new random backoff time.
• For each arriving packet belong to the interested
  flow, assign a random backoff time and put it in
  the competition with the rest un-sent packets
  from the moment it arrives at the base station.
• Packets older than a fixed threshold is
  considered to be timeout and discarded.
• Report whether and when the packets belong to the
  interested traffic should be send through the air
  interface.
• Data rate of packet transmission is the full
  bandwidth of WLAN.

24
Simulation Model

• Illustration of WLAN simulation process

Collision happened, both postponed
Timeout
DIFS
Data frame
Data frame
Data frame
user1
DIFS
BUSY
DIFS
BUSY
DIFS
Data frame
Data frame
user2
Data frame
DIFS
Backoff slot
Rescheduled backoff slot
DIFS
DIFS
DIFS
Data frame
Data frame
Interested flow
t
Sense busy when collision happened, resume the
backoff after sense the media free for DIFS
25
Simulation Model

• Illustration of WLAN - more microscopic view

CTS period
ACK period
SIFS
SIFS
SIFS
User data
Receiver
DIFS
Data frame
Data frame
user1
RTS period
DIFS
user2
Backoff slot
26
Simulation Model

• Illustration of WLAN - more microscopic view

CTS period
ACK period
SIFS
SIFS
SIFS
User data
Receiver
DIFS
Data frame
Data frame
user1
RTS period
DIFS
Data frame
user2
Backoff slot
If user2 starts RTS during the second period,
the first attempt wins and all the other fail.
Only the first attempt is scheduled in simulation.
27
Simulation Model

• Illustration of WLAN - more microscopic view

CTS period
ACK period
SIFS
SIFS
SIFS
User data
Receiver
DIFS
Data frame
Data frame
user1
RTS period
DIFS
user2
Backoff slot
Protected transmission period by MAC. All
competition in simulation is suspended by this
media busy period.
28
Simulation Model

• CDMA MAC algorithm
• Each station will acquire a PN code for spreading
  its signals.
• Stations do not coordinate with each other in
  sending packets.
• If too many transmissions happen in the same time
  period, the BER will increase during that period.
• Based on the increased BER, calculate whether the
  packet is received correctly or not by
  calculating the binomial distributed probability
  of error bits.

29
Simulation Model

• CDMA MAC algorithm (Contd)
• Additional concern
• In physical layer we assume perfect power
  control which eliminate the far-near impacts.
• In network layer call admission control scheme
  is considered to limit the number of users in the
  cell
• More we consider multi-service CDMA admission
  control.
• Admission control is calculated through

30
Simulation Model

• MAC competition model for CDMA systems
• For CDMA systems, set the time interval equal to
  the fixed frame time.
• Typical value 20ms in the CDMA systems
• The energy used in transmitting one bit is fixed
  for Eb, while the interference density is Io. If
  the number of users is N in the cell and the
  number of active users is Na, according to their
  ON/OFF model, the interference density is Na
  Eb/(W/R). Then the SNR can be represented as SNR
  Eb/ Io W/(RNa) where Na is a random variable.
• Classical CDMA analysis calculate Na based on the
  assumption that the ON/OFF model has Poisson
  distribution, which is not accurate.
• We want to model Na based on theorem1 in Will97.

31
Simulation Model

• MAC competition model for CDMA systems (Contd)
• According to the functions relating BER and
  SNRNich88, we can calculate the BER. The
  function for BPSK/QPSK is
• Then by calculating BER based on changed SNR, we
  can further decide whether the packet is
  corrupted or not based on the calculated BER and
  the threshold set by the error correction code.
• We only send out the un-corrupted packets via the
  air interface.

32
Simulation Model

• Illustration of CDMA competition

Bit errors calculated based on the SNRi
data
data
data
user1
data
data
data
user2
data
data
data
usern
t
SNR change points
SNR2
SNR1
SNR3
Bit errors occurred during all the transmission
period of a packet is accumulated to calculate
the total number of bit errors in the packet.
33
Simulation Model

• Illustration of CDMA competition(Contd)
• Models used to calculate the bit errors based on
  BER
• case 1 ---- OPNET model
• the number of bit errors in a segment of packet
  is based on the formula below

34
Simulation Model

• Illustration of CDMA competition(Contd)
• Models used to calculate the bit errors based on
  BER
• case 2 --------- Bernoulli experiments
• For each bit in the segment, do Bernoulli
  experiments on its correctness based on the
  distribution
• case 3 ---------- Binomial distributed random
  number
• The experiments can also be iterated to be in N
  samples, if the probability for a certain type of
  events to happen is p, then the number of events
  in N samples is a random variable conforming to
  binomial distribution.

35
Simulation Model

• Benefits of the above simulation model
• No need to implement each individual interfering
  node and the real MAC protocol modules.
• To implement each interfering node and MAC
  modules have the following drawbacks
• No realistic way to decide the traffic parameters
  for each mobile node except to get the real trace
  from the service providers.
• Eg. Too many parameters to decide LAN/WAN
  traffic, RTT, data/video flow, mean rate, etc.
• Cost of simulation is really expensive
• eg. To get the simulated aggregated trace by
  superposition of hundreds of ON/OFF individual
  source will need long time to compute on massive
  parallel machines. Will97

36
Simulation Model

• Benefits of the above simulation model
• The behavior of media sharing and competition is
  still valid through specific process models
  addressing for different MAC protocols.
• Easy to change the MAC process algorithm to
  address for a different MAC protocol without
  changing the simulation framework.

37
Simulation Model

• Urgent issues
• Parameters for three different classes of
  traffics
• in WLAN
• in CDMA
• Collision in WLAN
• do we allow multi-transmission in the cell? No,
  the amount of simultaneously transmissible
  traffic is extremely low.
• Traffic generation
• Whether the ratio among different classes is
  fixed/not?
• Based on the typical set of parameters for a
  traffic class and randomly generated ratio of
  offered load vs wireless link capacity in the
  cell, we can decide the number of users for each
  different class.

38
Simulation Results Affected by the Adjusted MAC
Model

• Metrics affected by this approach
• Only one metrics evaluating the interference of
  handover to new cell is affected
• Number of collisions / corrupted or discarded
  packets happened before the source start to
  regulate its rate for the new path.
• This metrics can be collected by the MAC
  competition process in the base stations.
• It addresses for the friendly behavior of the
  traffic control algorithm during the handovers.

39
Simulation Results Affected by the Adjusted MAC
Model

• Metrics affected by this approach (Contd)
• Number of collisions during handoff